feat(sentinel): restore JD01 cadence cronjob visibility

2026-07-01 06:43:30 +00:00
parent 622804f95d
commit c7ea34f253
12 changed files with 503 additions and 28 deletions
@@ -26,6 +26,14 @@ baselines:
        intervalMs: 600000
        heartbeatStaleSeconds: 900
        maxConcurrentRuns: 1
+        freshnessWarningMultiple: 2
+      observability:
+        otel:
+          enabled: true
+          serviceName: hwlab-web-probe-sentinel-${nodeLower}
+          tracesEndpoint: http://otel-collector.platform-infra.svc.cluster.local:4318/v1/traces
+          sampler: parentbased_traceidratio
+          samplerArg: "1"
    scheduler15m: &scheduler-15m
      intervalMs: 900000
      heartbeatStaleSeconds: 900
@@ -229,8 +237,15 @@ nodes:
            maxSeconds: 360
            serviceUnavailablePolicy: structured-failure
            cadenceScheduler:
-              enabled: false
-              reason: cicd-health-endpoint-only
+              enabled: true
+              reason: k8s-native-periodic-quick-verify
+              concurrencyPolicy: Forbid
+              startingDeadlineSeconds: 600
+              successfulJobsHistoryLimit: 3
+              failedJobsHistoryLimit: 5
+              activeDeadlineSlackSeconds: 60
+              ttlSecondsAfterFinished: 86400
+              backoffLimit: 0
        secrets:
          sources:
            - <<: *jd01-bootstrap-source
@@ -28,6 +28,7 @@ sentinel:
      intervalMs: 600000
      heartbeatStaleSeconds: 900
      maxConcurrentRuns: 1
+      freshnessWarningMultiple: 2
    sqlite:
      path: /var/lib/web-probe-sentinel-auth-switch/index.sqlite
      busyTimeoutMs: 2000
@@ -28,6 +28,7 @@ sentinel:
      intervalMs: 600000
      heartbeatStaleSeconds: 900
      maxConcurrentRuns: 1
+      freshnessWarningMultiple: 2
    sqlite:
      path: /var/lib/web-probe-sentinel-dsflash/index.sqlite
      busyTimeoutMs: 2000
@@ -28,6 +28,7 @@ sentinel:
      intervalMs: 600000
      heartbeatStaleSeconds: 900
      maxConcurrentRuns: 1
+      freshnessWarningMultiple: 2
    sqlite:
      path: /var/lib/web-probe-sentinel/index.sqlite
      busyTimeoutMs: 2000
@@ -29,6 +29,7 @@ sentinel:
      intervalMs: 600000
      heartbeatStaleSeconds: 900
      maxConcurrentRuns: 1
+      freshnessWarningMultiple: 2
    sqlite:
      path: /var/lib/web-probe-sentinel-fake-echo/index.sqlite
      busyTimeoutMs: 2000
@@ -28,6 +28,7 @@ sentinel:
      intervalMs: 900000
      heartbeatStaleSeconds: 900
      maxConcurrentRuns: 1
+      freshnessWarningMultiple: 2
    sqlite:
      path: /var/lib/web-probe-sentinel-mdtodo/index.sqlite
      busyTimeoutMs: 2000